right to physical security, would be so absolute or unyielding as to trump any other human rights". 70 4.2 SOCIAL MEDIA ALGORITHMS AND HUMAN RIGHTS DUE DILIGENCE The international human rights standards governing responsible business conduct were formulated before the human rights implications of artificial intelligence were widely appreciated – and this has created challenges for those seeking to hold Big Tech companies accountable for their human rights impacts. This accountability gap is exacerbated by the fact that Big Tech’s use of algorithms and artificial intelligence has been marked by an absence of transparency regarding how these systems work, and because the inherent complexity of such technologies can require specialized knowledge to unpack and analyse. WHAT IS A SOCIAL MEDIA ALGORITHM? Social media algorithms are pieces of computer code comprising a set of rules, which enable automated decisions and recommendations to be made on a social media platform. These algorithmic systems shape a user’s experience on the platform and determine what information they see - for example, what post someone sees at the top of their Facebook News Feed. Social media platforms use a wide range of algorithms in the delivery of their services.71 These algorithms rely on artificial intelligence and machine learning technology, meaning that they operate under varying levels of human oversight. There are two major categories of algorithms which are particularly relevant to this report: 1. CONTENT-SHAPING ALGORITHMS Content-shaping algorithms are designed to serve a user 'relevant' content (relevancy that is inferred by the companies on the basis of collected personal data). These algorithms are continuously trained on vast amounts of user data to serve many different purposes, such as ad targeting and delivery, serving search results, recommending new content, and prompting users to create new content and engage with existing content. To do this, the systems “optimize” to best deliver a specific outcome using highly complex and iterative algorithmic processes that draw correlations and inferences from user data. Examples of these content-shaping algorithms include the Facebook News Feed and ranking algorithms, which decide what users see on their news feed and in what order, and its recommender algorithms, which decide what content is recommended and auto-played for specific users. 2. CONTENT MODERATION ALGORITHMS Content moderation algorithms are increasingly used to automate social media platforms’ detection, demotion, and deletion of content which violates platform rules.72 The algorithms are partly trained by human staff, who classify and label millions of pieces of content in order to train the algorithms on how to automatically detect violating content. Although content moderation was previously conducted primarily by human moderators, it is increasingly automated. Content moderation algorithms can also take automated decisions in relation to content, including by ‘demoting’ content in the news feed rankings, or by removing content entirely. In recent years, international standards have begun to evolve to accommodate these innovations and industries. The UN Office of the High Commissioner for Human Rights (OHCHR)’s B-Tech Project has affirmed that tech companies’ due diligence must also include addressing situations in which "business model-driven practices and design decisions create or exacerbate human rights risks", as well as an analysis Working Group on the issue of human rights and transnational corporations and other business enterprises, “Business, human rights and conflict-affected regions: towards heightened action”, 21 July 2020, A/75/212, para. 99. 70 In 2019, New America’s Open Technology Institute published a series of reports setting out how automated tools are being used by internet platforms to shape the content we see and influence how this content is delivered to us. See Spandana Singh, Open Technology Institute, Everything in Moderation: An Analysis of How Internet Platforms Are Using Artificial Intelligence to Moderate User-Generated Content, 2019 newamerica.org/oti/reports/everything-moderation-analysis-how-internet-platforms-are-using-artificial-intelligence-moderateuser-generated-content 71 ‘Content moderation’ refers to social media platforms’ oversight and enforcement of platform rules in relation to permissible and prohibited forms of expression - for Meta, these rules are known as ‘Community Standards’ (see Section 5.4 below). For additional information, see: “A Tale of Two Algorithms” in Ranking Digital Rights, It's Not Just the Content, It's the Business Model: Democracy’s Online Speech Challenge, 17 Mach 2020, available at newamerica.org/oti/reports/its-not-just-content-its-business-model 72 THE SOCIAL ATROCITY META AND THE RIGHT TO REMEDY FOR THE ROHINGYA Amnesty International 20